Notes on coding agents and the pages they leave behind, on moving between notes apps, and on the plain mechanics of keeping what you think. Written and published from a Sumibako vault.
ChatGPT says you've reached the maximum length for this conversation. How to carry on
What ChatGPT's maximum conversation length is, why chats hit it sooner since September 2026, and a handoff note template so a new chat carries on.
Prashant Shaw
Founder, Sumibako
8 min read
You have spent a week in one ChatGPT conversation. It knows the project, the decisions, and the three things you tried on Tuesday that did not work. Then it stops: "You've reached the maximum length for this conversation, but you can keep talking by starting a new chat." The new chat is eager, polite and knows nothing whatsoever, like a golden retriever on its first day at your company.
Since late September there is a second version of the same wall, called "Chat memory full", and a lot of people are hitting one or the other much sooner than they used to. Below: what OpenAI says about the limit (not much), what changed, and what works, from the new button to a handoff note you write yourself and a project that outlives any one chat. I build a notes app, which turns up once near the end. Everything is linked and was checked on 18 October 2026.
"You've reached the maximum length for this conversation. You can start a new chat to continue."
Every one of these comes from OpenAI's community forum, quoted by people who met it. I could not find any of them on OpenAI's help pages or in its release notes. The April warning was written down from memory by the person who saw it ("It goes something like this"), so trust its first sentence more than the rest of what they recalled.
They are the same wall with different paint. When a user posted the "Chat memory full" notice, OpenAI's support account replied on 2 October that it "indicates that the conversation reached its available memory/context limit". It is not your saved memory, the list of facts ChatGPT keeps about you, and clearing that list will not reopen the chat. That one has a "memory full" of its own, which I wrote up separately.
đ How long is too long?
OpenAI does not publish a length for a conversation. What its pricing page does publish is the context window: how much text, in tokens, the model can consider at once.
Plan
Instant
Thinking
Free
27K
Varies
Go
54K
256K
Plus
54K
256K
Pro
128K
400K
Under the table sits a footnote worth reading: "The portion available for user input is smaller than the total window, as space is also used for system instructions (including tools and personality), memories (if enabled), and internal processing". Your conversation shares the room with your memories, the tools and the model's own thinking, and it is not the one holding the lease. The rest of what OpenAI publishes is in ChatGPT's limits, plan by plan.
The window is not the wall, though. People have long described a chat that outgrows its window as quietly losing track of its early messages while carrying on. The wall is different: the chat stops taking messages at all. Where it sits is not written down anywhere I could find, and reports in the forum run from a month of daily use to fewer than 20 prompts.
đ Why it got worse in September
If this happened to you recently, you are in a crowd. From mid-September the forum filled with chats reaching the limit far sooner than before: one on Plus "after fewer than approximately 20 user prompts", a fiction writer also on Plus getting about half the length they used to, and replies from people on Pro and Business saying the same. From about 30 September came "Chat memory full", which people report after as few as 10 to 30 messages, some of them on the free plan.
OpenAI's support account wrote on 4 October: "We're looking into the issue and will share an update here if there are any developments." A week later it asked anyone still affected to contact support. That is the whole of the official record I could find. Nobody at OpenAI has said the limit changed, and forum posts are reports, not measurements. The useful conclusion is a narrow one: by these accounts, paying more did not make the wall go away, so plan for it on any plan.
đ The new button
The "Chat memory full" notice comes with a way out, which OpenAI's support account calls "Continue in a new chat": a new conversation that starts with an automatic summary of the old one. It beats a blank page by a distance. Two things to know before you trust it:
You did not choose what went into the summary. ChatGPT did, at the moment it had run out of room, which is not when anyone does their best writing.
Inside projects it has a reported bug. One user saw "Continued from" after every message, with the history vanishing until a refresh, and support replied that there is "a known issue with some Project conversations where branching can make parts of the chat history look like they're missing."
So press it, then test it. Ask the new chat to tell you in two sentences where things stand. If its answer is wrong, the summary is wrong, and you found out in ten seconds instead of an afternoon.
đ¤ A handoff note you write yourself
The summary is ChatGPT's account of the chat. A handoff note is yours: what was decided, what was ruled out, and what happens next. I wrote the long version for Claude's equivalent limit, and it works the same way here. The rules, briefly:
Ask for it before the wall. The "nearing its limit" warning is your cue. A full chat still opens and can be read, but it will not take another message, so it cannot write you anything.
Ask for the work itself, in full: the current draft, code or table. A description of a draft is not a draft.
Include what was ruled out, or the new chat will cheerfully suggest Tuesday's idea again.
Send this in the old chat:
Then open the new chat with the note and this line:
If the old chat has already closed, read it and copy the latest version of the work out by hand. It is tedious, and it is still faster than explaining Tuesday again.
đī¸ Keep long work in a project
The better fix is to stop betting everything on one chat. A project keeps chats, files and instructions together, and every plan has them, Free included. Three steps:
1ī¸âŖ Move the long chat in. OpenAI's instructions: "drag a chat onto your project, or open a chat's menu and choose Move to project".
2ī¸âŖ Put the handoff note in the project's files or instructions, so every new chat there starts with it. Free allows 5 files, Go and Plus 25, Pro 40.
3ī¸âŖ Start the next chat inside the project. With memory switched on, chats in a project "can draw on other conversations within the same project".
Two habits help as well. Take side questions to a branch: hover over a message, open More actions and choose "Branch in new chat", per the release notes. A branch starts from the message you pick, so it keeps tangents out of the main chat, but it is no escape from a chat that is already full. And attach long documents rather than pasting them. Above 10,000 characters ChatGPT now does it for you, which, per the release notes, "helps prevent large pastes from consuming your entire context window."
đ Where to keep the note
Where
Who can use it
The catch
Pasted as the first message
Everyone
You carry it by hand, every time
In a project's files
Every plan
It only helps chats inside that project
A Page in ChatGPT's Space
Pro, Business and Enterprise, since 29 September (release notes)
Not on Free, Go or Plus
A notes app, through a plugin
Business, Enterprise and Edu to write. Pro to read
Not on Free, Go or Plus
A notes app, pasted by hand
Everyone, in any AI app
You are the plugin
The last two rows are where I come in, so adjust your trust accordingly. Sumibako is a notes app that AI apps can write pages into. On ChatGPT that needs a plan that allows custom servers: OpenAI's own page says "Full MCP is only available to Business and Enterprise/Edu users, currently", and Pro can connect one "with read/fetch permissions". The setup is here. Claude can read and write it on every plan, Free included, through a custom connector. For a handoff note, that means one page, updated rather than copied each time, that you can correct by hand and that belongs to no single chat or AI app. The version it overwrote is kept for 14 days.
What it is not: it has no AI of its own, so it will not write the note for you, and it saves the page you give it, not the transcript.
The wall was always there. What changed in September is how soon you meet it, so write things down while the chat can still hold a pen.
Eight eMudhra keys were integers under 256, one of them 4. Why browsers blocked them in hours, why your Docker image did not, and what to check by 27 October.
10 min read
Write a handoff note for a new conversation that has not seen this one.
Use these headings, in this order:
Goal: what we are making and for whom, in two sentences.
Decided: each decision, with its reason in a few words.
Ruled out: what we tried or rejected, and why.
Current state: the latest version of the work, in full, not described.
Open questions: what is unresolved, most important first.
Next step: the one thing to do next.
How I like to work: tone, format and limits I have asked for.
Leave out anything that no longer matters.
This is a handoff note from an earlier conversation. Tell me in two
sentences where we are, then do the next step.
